I mean, the fact the same prompts give you images that look like the ones in the game is the problem, mate. The images in the game are not of very high quality. There's also the fact that you're using tokens to generate images - 2 gb of AI images is nothing, I'd have Terabytes if I wasn't religious about deleting all but the very best.
Mine is able to do lewds too, but that's not really the issue. It doesn't surprise me that you're having difficulty with non-standing lewds; The thing I was trying to emphasize was how to change the style of the image.
The style makes a pretty significant difference, imo.
One thing I've noticed about online generators - and I do try to keep up - is that they're a lot more restrictive, particularly in which tags have effect. Every site is subtly different and most of them restrict tags that are... Questionable.
Did you use the huggingface UI, or did you use the huggingface AI, which is online? There is actually a difference, and it's very profound.
As you can see, I have little issue generating lewd pictures in which the girl isn't standing:
The pictures have the same art because i'm using the SAME art style. And don't have very high quality because rpg maker works bad with 2k pictures and .webm reduce the quality a LOT. Just rpgm maker limitations.
Make other art styles is easy, but i choose only one style to make the game more consistent, and not have like 20 art styles diferents.
If you think that this art is worst than one you send me, you have a serius problem. Yours pictures looks old, and have ai problems, the pussy of the girl you send me looks creepy.
And the pictures of my game are not just the girl "STANDING" have a lot more posses.
I used huggingface UI, on local, and is trash.
The girl with the tentacles was my error, not the ai model error, I should have ignored the image, not introduced it into the game, it was a human error.
In that case, I'd like to give a few tips. I'll include examples in spoilers too. For reference, I am using an NovelAI build - you should be able to get BETTER and more consistent images off of newer AIs.
Tip 1: If you're not using the webui from huggingface, you should. It's complicated to set up, but once you manage it you have access to very good AIs directly on your computer, without any need to go online. That means you can much more easily pick and choose which to use, or have uncensored versions happy to generate whatever you need. The Webui also includes options for inpainting and using an image as a prompt, both of which are useful. Once it's set up properly, it should look like this:
Tip 1.5: If you do use the webui above, although the default options work, I recommend the following:
-Set Sampling Steps to somewhere between 28-32. This strikes a good balance between the pictures taking ages to generate, and them looking like garbage
-CFG scale should be around 11. Too high and you get weird outlines on your images. Too low and they become incoherent.
-Width and Height can be played with, but I find that 640x640 gives the AI more room and that helps it place things correctly. If you change it, try reducing the other side by the same amount; the shape of the image will change what the AI tries to do. Images wider than they are tall will make it want to do a landscape, while images taller than they are wide will make it want to draw a portrait of one person. This is very useful!
-Generate in small batches while you're dialing in on the right tags to use, then swap to 3 batches of 3, or 4 batches of 4. Then select your best.
Tip 2: Use the Traditional Media tag. On it's own it doesn't do a whole lot - although images will look subtly better. Use the Traditional Media tag in addition with a type of medium normal art is done in. For example, watercolors, colored pencils, or charcoal. I'm a particularly big fan of Charcoal. Woodcuts, 3D, it all works though. Sometimes, not very well, but often, better than nothing at all. Here's a couple examples:
Something a lot of AI sites don't make entirely obvious is how to whitelist, blacklist, and emphasize things. These are very useful for producing images that look legit.
As you can see, the more emphasis, the more the AI tries to put it in, up to - and beyond - the point where it distorts the image. Meanwhile, de-emphasizing a tag makes it less prominent, and more subtle, until it goes away entirely.
You'll also notice I'm using stuff like "masterwork" and "best quality" in the whitelists, and I'm blacklisting common problems like "poor quality" or "missing fingers". Doing this helps keep the quality consistently mediocre instead of godawful.
I'm also doing large batches with tons of generations. It took almost an hour to make all the pics for this post, most of it waiting for generations. This will be you. The AI will NEVER consistently make high-quality generations every time, it's just not good enough yet. You need to spend the time generating trash so you can sift through and find the best stuff.
I encourage anyone else in this thread to look through the images and see if you have a preference towards media type. It might help the dev choose which he'll use.
No, I skipped most of the technical details of moving the AI onto your computer. It's not a tutorial so much as advice on how to prompt the AI for a more distinct aesthetic.
I think finding a unique and stand-out aesthetic is important for AI-generated content. A lot of people loathe the default look and it is kinda... Meh. Standing out helps get people interested, which can have compounding effects on how large the fanbase gets (and how much money you can make, if that's your design)
The pictures have the same art because i'm using the SAME art style. And don't have very high quality because rpg maker works bad with 2k pictures and .webm reduce the quality a LOT. Just rpgm maker limitations.
Make other art styles is easy, but i choose only one style to make the game more consistent, and not have like 20 art styles diferents.
If you think that this art is worst than one you send me, you have a serius problem. Yours pictures looks old, and have ai problems, the pussy of the girl you send me looks creepy.
And the pictures of my game are not just the girl "STANDING" have a lot more posses.
I used huggingface UI, on local, and is trash.
The girl with the tentacles was my error, not the ai model error, I should have ignored the image, not introduced it into the game, it was a human error.